Key points
On August 19, 2026, IBM announced at Yorktown Heights, NY that it had connected two box-shaped modular cryogenic units and cooled them to 15 millikelvin (mK)—about 273.135°C below zero, roughly 180 times colder than the cosmic microwave background. This marks the retirement of the cylindrical 'golden chandelier' dilution refrigerator used since IBM Quantum System One (2019), and the first hardware foundation for the Starling 2029 roadmap: 200 logical qubits, 10,000 physical qubits, and 100 million fault-tolerant operations.
The new modular cryogenic platform
- Vacuum chamber volume: 2.75 m³
- Wiring surface area: 0.53 m²
- Wiring capacity: 12× Quantum System One
- Modules connect via L-coupler superconducting cables, transmitting microwave photons over ~1 meter between units
- Room temperature → 4 K (liquid helium stage) in under 5 days
- Further cooldown to below 15 mK shortly after
- With a simulated 30 µW heat load (future processor + wiring + electronics), the system held stable at 23 mK
- First demonstrated in 2024 as a meter-scale cryogenic superconducting cable
- Current cross-module two-qubit gate fidelity: 99.3%; target: 99.9% (the fault-tolerance threshold)
- The 0.6-point gap is decisive: below 99.9%, error correction demands far more physical qubits
- Quantum System Two's three core components (control electronics, cryogenics, processor modules) are now independently testable, compressing iteration cycles from 12–18 months to 3–6 months
- Modules are factory-prefabricated, transportable, and field-assembled—cryogenics becomes a standard product, not a bespoke project
- Starling will deploy at IBM's Poughkeepsie, NY facility as a quantum-classical hybrid: quantum modules compute, FPGA decoders correct errors in real time, classical HPC handles pre/post-processing
The old paradigm hit a scaling ceiling: wiring density, heat dissipation, and single-chip limitations made expansion beyond ~1,000 qubits impractical in one cryostat.
Cooling performance
This validates a new engineering paradigm: cool first, install chips later—modules are cryogenically tested in the factory, shipped, assembled on-site, and only then fitted with quantum processors, reducing transport risk to calibrated qubits.
L-coupler cross-module entanglement
Starling roadmap milestones
| Time | Milestone | |---|---| | 2026 Q4 | Nighthawk processor installed in modules | | 2027 | 1,000+ programmable physical qubits; L-coupler fidelity → 99.9% | | 2028 | 12-module fault-tolerant subsystem | | 2029 | Starling: 200 logical qubits, 100 million fault-tolerant operations, cloud access |Error correction: qLDPC at 12:1
IBM's qLDPC codes (developed 2024) require roughly 12 physical qubits per logical qubit—versus ~100:1 for Google's surface code and ~9:1 for Amazon's Ocelot—compressing hardware overhead near theoretical limits. IBM also demonstrated FPGA-based real-time decoding, enabling error correction during computation rather than between runs, a key reason third parties (e.g., Riverlane) view the roadmap as credible.Industrialization signals
Competitive context
| Dimension | IBM | Quantinuum | Google | |---|---|---|---| | Qubit type | Superconducting transmon | Trapped ion | Superconducting transmon | | Current qubits | Heron 156 → 1,000+ cross-module | Helios 98 | Sycamore 70 | | Error correction | qLDPC (12:1) | High-fidelity QEC | Surface code (~100:1) | | Interconnect | L-coupler meter-scale cables | Shared ion traps | No public scheme | | Fault tolerance target | 2029 Starling | 2029–2030 Helios 4 | 2029–2030 Willow 3 |What to watch in the next 6–12 months
1. Nighthawk installation on schedule (2026 Q4) 2. L-coupler fidelity breaking 99.9% (2027 Q1) 3. 1,000+ physical qubit multi-module system (2027 Q2) 4. qLDPC stability at scale (2027 Q3) 5. 12-module subsystem integration (2028)Any delay gives Quantinuum, Google, or trapped-ion challengers an opening—the 2029 quantum race effectively began in 2026 Q4.
---